I don't know if you've found a solution to this, but I've spent the last few days trying to compile different options because we have a very similar issue in our group. There seem to be two options to get around the getframe issue:
1) Use "print" function with the -RGBImage flag, returns a cdata array that is equivalent to frame2im(getframe)
2) Use Xvfb (X virtual frame buffer) to load a virtual X11 context for matlab to run in on your compute nodes (see this solution out of MSU: https://wiki.hpcc.msu.edu/pages/viewpage.action?pageId=19595894), requires Xvfb to be installed.
Here's some basic timings to get an idea how these vary (only timing the saving png file phase).
Test system: plot(rand(5000));
- no x11 window (blank opengl) - saveas: 580s
- no x11 window (blank opengl) - print RGB: 348s
- X11 Xvfb (software opengl) - getframe: 107s
- X11 actual window (software opengl) - getframe: 33s
So Xvfb is 3-4x worse than an actual X11 window, but still quite a bit faster than just calling saveas. Print is pretty costly overall, but it does allow some speedup if you don't have access to Xvfb. As a side note: for Xvfb if you compute values and want them printed out to read in a log file, make sure to make a diary as all the matlab output will be pushed into that virtual X11 window (which will just disappear at the end).
Hope this helped,
Andy Long
