ImageJ is bizarre in its haphazard provisioning of 'undo'. This is a shock after coming from Gimp with its unlimited depth of undo (plus direct navigation to the last acceptable versions). Sometimes ImageJ has one level, sometimes none at all -- and only sometimes is that preceded by a warning. It doesn't matter how much RAM is allocated to it.
For workarounds, I tend to do short sequences of commands, then save or more commonly exit to Gimp via Copy to System Clipboard for polishing steps. That way, no harm done by no undo -- just close the file, re-open and re-do. Alternatively, I make a duplicate of the working file before experimenting on it with a dodgy new command. Some menu items offer this as a checkbox.
ImageJ is also odd in that it is hard to say when it has committed to a step vs simply presented you with a preview not labelled as such.
It also provides no choices within the set of open windows. For example, it will try make a stack out of everything open, there is no way of selecting a subset. There is no way of re-ordering files in a stack -- that maymake no sense for a time series but does in making attractive colors by shuffling RGB channels. Here the files must be sequentially uploaded in the desired final order.
For these reasons -- and a memory leak that can drain 16 GB of RAM, leaving other applications with nothing -- I tend to shut it down frequently and get a clean start. The programmers claim this is not a memory leak per se -- I guess it all depends on what you mean by 'per se'. Excellent resource: http://imagej.net/Frequently_Asked_Questions
While I am yet to use 2% of the plugins, it's worth noting that biomedical film, slicing, ROI movement and gel enhancements, as well as their analysis, are very similar to improving poor contrast images encountered in Greenland, the Arctic and Antarctica and in making time series of ice and glacier movements. Both groups of researchers are stuck with a certain level of resolution and have to make the most of it.
One more thing I use a lot: the vast range of file types that can be imported. Here is a big difference between a scientific tilt (ImageJ) and artistic (Gimp) -- the former are very keen to import raw data as image.
One import command that is really under-used on these forums is "Extract images from PDF..." which pulls out the full author-submitted resolution of journal pictures. You cannot get at these by simply enlarging a journal pdf in a web browser or pdf reader as these just provide quickie linear interpolation that loses resolution upon screen capture. If the authors only submitted a cheesy resolution, try imageraider.com to locate a better resolution image elsewhere on the internet.