First a few questions to take better guesses/clarify:
-How far off when you say "angled slightly?" As in: if you were to mark your steering wheel centerline at the 12 o'clock position, with a matching mark on the dash, when you are experiencing it being off, how far in mm or inches is it off to either side? You can use some tape on dash/wheel to help gauge it if you need to.
-How long is "after a while?" 20 seconds? 20 minutes? 20 parsecs?
-You said it also does it after turning off the car and starting again... are you just stating that once it immediately starts doing it, you have tried turning it off/restarting, and it continues to do it for the same amount of time as it would if you hadn't restarted? It was worded a bit strange just checking. Frankly restarting shouldn't have anything to do with anything suspension related, our racks are electrically assisted but they don't have any voodoo inside like BMWs with variable ratio racks.
A few potential wild @&% guesses/random thoughts:
-Toe in can have a somewhat similar effect(numbs the turn-in, as a result requires more steering input initially... which also translates to it not being as "precise" right around the straight ahead position. Checking the amount of toe in/out isn't a bad idea. Going from -0.20 total toe to +0.20 total toe can have a huge effect on steering feel/precision in the straight ahead position. Can be checked easily at home as per here:
http://www.mazdas247.com/forum/show...ber-bolt-install-tip-string-alignment-at-home
-Maybe steering rack bushings? Normally they would cause some noise or it would be blatantly obvious stuff is shifting around at other times, but it might be a good idea to see if you can force it back and forth by hand where it mounts to the subframe. A rack will move a pretty noticeable amount under heavy loads but I've never experienced a slow return back to center myself. While turning left the rack would theoretically move left, causing the knuckles to both turn right a small degree, meaning requiring another left input to keep it tracking straight once done turning.
-You can rule out rear suspension beam shifting - under a left turn, it would force itself to point to the right(positive thrust angle), requiring counter-steering in the right direction to keep straight afterwards.
-Did you ever have the control arms loosened during the coilover install process? If so for the front bushing(with the bolt that inserts horizontally), was it torqued with the weight of the car on the ground and at ride height? This will induce bushing binding at ride height if it's torqued with the car in the air... I could MAYBE see how the bushings having preload on them could potentially cause a temporary change in ride height(even 5mm has a HUGE effect on toe)... basically the inside body will sit higher for a short amount of time after being unweighted in hard cornering, and driving around it will eventually settle back to normal as it twists and gets stressed back into position(where it's binding at all times). If torqued in the air, the control arm wants to naturally be in the "down" position.
-When turning left, under load, the upper strut mount would want to move outwards(losing camber). If it is a coilover that is reusing stock style rubber mounts, maybe possible they're flexing and shifting around... Upper strut moving outward on the outside(right) only would yield a loss of toe(more toe out), so would require a left input to counter steer assuming it's sticking after a turn. If it's a metal upper mount you can probably discount this entirely.
-Kind of going off the control arm binding theory... know what else can affect ride height? Shock damping. If there is a metric ton of rebound damping just maybe it's giving you a jacking down effect like some autocrossers use to artificially lower ride height during cornering on "street" or "stock" class vehicles. Lots of damping with not enough spring will cause this. If it takes more than about a minute of driving for it to "straighten out" again I'd rule this out. The effects of the car jacking down would go away after 10 to 15 seconds at the most. You can try turning the damping down some to try and rule it out.
-Lower control arm bushings deforming under load on a left turn on the outside wheel would result in an increase of toe(in) on that wheel... meaning to counter it would require a right input. Opposite of your symptoms.
-Check the steering shaft from column to rack at all? I'm assuming you'd hear a clunking if there were play, but play here would cause the correction in the same direction of the turn as you describe, though nothing would give a "delayed" reaction since it's all metal. It can probably be ruled out but is worth checking.
-Or you're just imagining things. Is this always on flat roads? One roads with a crown to them they will always run a bit off center one way or the other due to the car having weight shift toward the lower side, meaning you're trading toe in on one side for toe out on the other or visa versa a tiny amount, meaning the wheel will be slightly off center and pulling would be slim to negligible anyway being a FWD car(in my experience it takes a big screw up in alignment, caster specifically, to actually cause a real pull, or defective tire).
-Out of curiosity have you tried dicking with tire pressures at all just to see if it changes or stays the same? Try airing at 30psi, then try again at 40-45psi just to see if there's a perceivable difference in how long it lasts or how far off it actually goes.
Answer the questions above and maybe I'll try another crack at diagnosing a car from the other end of the world (hi)