Paul Stoffregen: Is that 127KB a uploading performance test or an application you have?
I just clicked the Upload button in Arduino and watched a clock window on my computer, which admittedly isn't highly accurate, especially since I'm looking at the LED on the board and then I look over at the screen to see how many seconds elapsed.
I just tried it again, using a digital kitchen timer (with 1 second precision), where I click the Upload button with 1 hand and the timer start button with the other hand. The sketch I'm loading is File > Examples > Basics > Blink with the pin changed to 6 (since Teensy++ has a LED on pin 6), and I set the Tools > USB Type to "Disk(internal)" (which causes all unused flash memory to become a tiny USB disk drive, creating a maximum size upload of 127K). After clicking Upload, I stare at the LED with my finger on the stop button, which I press the instant the LED lights again. The first second or so it continues to blink, while the PC compiles code, and then stops blinking during the actual upload. It lights up again when the freshly-uploaded code begins running. That occurs right after the timer advances from 4 to 5 seconds, so it's between 5 to 6 seconds, but much closer to 5 than 6. Sorry, I don't own a sub-second stopwatch. I could build something, but I'm not going to. This is as accurate as I'm willing to measure for the sake of this conversation.
Again, the 5 second is from the moment the Upload button is clicked until the LED starts blinking again, so it includes the compile step and any overhead, and perhaps a slight error of human reaction time.
I stopped when the LED lights, but since Teensy is a native USB device which has just rebooted, the PC takes a moment to detect the newly-connected USB device. I did NOT measure the extra time (on the scale of a second or two) until Linux fully enumerates the USB device and the kernel and udev deamon create new device files, and a moment later a new window appears showing the files (since in this test Teensy was configured to be a USB disk).
While those 5 seconds do include the compiling, when using Teensyduino, Arduino is patched with a speedup which avoids recompiling previously compiled files. I contributed this code to Arduino and it has become part of 1.0.1-rc2, so you can grab that version for a small speedup in the compile step. I repeated this test, but changed boards back and forth before clicking upload (changing Tools > Board causes a full recompile). It measured 6 seconds, and the timer had been on 6 and might have been just about to advance to 7... so the full recompile adds about 1.5 seconds on my machine (which is Ubuntu 10.10, 32 bit, on a 2.8 GHz core2 processor, 2009 vintage). The full recompile for Teensy probably takes longer than Arduino Uno, since it's recompiling a complex USB stack as well as all the usual Arduino functionality.
Anyway, that's exactly how I measured 5 seconds for 127K upload, since you asked....
Trust me, one of the first comments I regularly hear from long-time Arduino users when they try Teensy is how much faster the upload goes. It's native USB with a specially written application, so it ought to be fast! The speed varies slightly on each operating system, but it truly does upload very quickly compared to using AVRDUDE on conventional boards.