Yeap, I decided to play a little with the debugger again... It is strange as sometimes just a 'little' profiling can speed things a lot more...
Now, this time, I did 2 improvemens:
1. The way that the main 'dispatch' was being handled: every time a new dispatch was called, I got the file information again from the frame (and had to do a couple of os.path calls), but in the end, the overhead for that was huge, so, now I 'cache' the context so that the filename and the base of the filename are stored in a cache.
2. The way that the shell was communicating with the shell was in the 'main' loop, for each call... now it is in a thread (I thought that this was going to be the major speed up, but number 1 was much more effective at it).
That will be available at release 0.9.8.8