Android :: HTC Specific Open GLES 1 Frame Rate / Can't Get It Right

I am trying to get a quite simple openGL ES 1 program run a smooth solid 60fps on a couple devices out there, and I get stuck on HTC desire. The phone itself is quick, snappy, powerful, and overall a breeze to use ; however, I can't seem to display anything fullscreen at 60fps with OpenGL. After getting stuck for a long time with my app, I decided to make a test app with code right out the sample code from the documentation.Here is what I am doing. Simple initialization code with GLSurfaceView. I have three versions of onDrawFrame, all dead simple. One is empty. One contains only glClear. One contains just enough state to only draw a fullscreen quad. Trace times before, and after. There is no view other than my GLSurfaceView in my program. I can't explain the times I get.In all cases, the onDrawFrame function itself always finishes under 2ms. But very often, onDrawFrame does not get called again before 30~40ms, dropping my frame rate all the way to 30fps or less. I get around 50fps with an empty onDrawFrame, 45 with glClear and 35 with a quad.
The same code runs at 60 fps on the HTC Magic, on the Samsung Galaxy S, on the Sharp ISO1. Sony Experia X10 caps at a solid 30fps because of its screen. I have been doing much more complicated scenes at a solid 60fps on the HTC Magic which is very underpowered compared to the Desire. I don't have a Nexus One in handy to test. Sure, I except buffer swapping to block for a couple milliseconds. But it just jumps over frames all the time.

Trying to find out what the phone is doing outside of the onDrawFrame handler, I tried to use Debug.startMethodTracing. There is no way I can get the trace to reflect the actual time the phone spends out of the loop. At the end of onDrawFrame, I use startMethodTracing then save the current time (SystemClock.uptimeMillis) in a variable. At the start of the next one I Log.e the time difference since the function last exited, and stopMethodTracing. This will get called over and over so I arrange for stopping once I get a trace for an iteration with a 40+ ms pause. The time scale on the resulting trace is under 2ms time, as if the system was spending 38ms outside of my program, I tried a lot of things. Enumerating EGL configs and try them all one after the other. Just to see if it changed anything, I switched to a render when dirty scheme requesting a redraw at each frame. To no avail. Whatever I do, the expected gap of 14~16ms to swap buffers will take 30+ms around half the time, and no matter what I do it seems like the device is waiting for two screen refreshes. ps on the device shows my application at around 10% cPU, and System_server at 35%. Of course I also tried the obvious, killing other processes, rebooting the device... I always get the same exact result.