You have two choices to get consistent HSPA+ performance on an imported device.
1) If you're on a MEdia Net or Data for Non-Smartphones plan you need a SIM issued that reflects one of those plans. Being on another plan and using the Cingular APNs won't get you a consistent result. Being on an iPhone plan will always get you an inconsistent result.
2) If you're on a 4G plan (non-LTE) and using what everyone refers to as the "Phone" APN you need a SIM card that reflects that plan. You can't use the phone APN if you're on a non-4G plan or the Cingular APN if you're on a 4G plan. The result will always be inconsistent and will vary as you roam across different AT&T regions. I have zero knowledge of what being on an LTE plan and using an associated SIM in a non-LTE device does but Iím guessing itís not too good.
The only difference between the two plans is the servers you're routed through to reach the network. Upload and download speeds on HSPA+ are the same; AT&T only has one HSPA/HSPA+ network. The 4G servers are what AT&T uses to throttle and measure data consumption and also the reason they and the "Phone" APN were introduced with the launch of the Infuse and Inspire. The benefit of you being on them is AT&T's, not yours. For now, there's no soft cap on the Cingular servers because AT&T doesn't have a mechanism in place to police them. And the first 5-7 numbers of the IMEI are standardized and identify the make and model of a device. So if AT&T isn't detecting what types of phones are running on their network it's only because they haven't decided to yet. If hundreds of thousands of imported smartphones are suddenly chewing up GB's of non-smartphone data poking the bear will eventually get its attention.
I'm on my fourth imported device and learned both from trial by fire and a pretty cool Premier account rep how the system works and how to work around it. Anything other than items one and two above are a crap shoot and won't get you consistent performance no matter what you do with your APN settings or SIM cards.