Let’s see some things you can do to avoid problems or even crashes when trying to print large, detailed, heavy models. Detailed models contain a lot of vertices and Slic3r will use more memory to handle them; sometimes such needed memory is more than your physical system memory.
First off, make sure you’re using Slic3r 0.9.10 or newer, as their memory usage was reduced enormously.
The first thing you can to is to reduce the number of vertices/facts in your model: if your file has more than 300k facets, it’s very likely way more detailed than you can achieve using your printer. You can then use Meshlab to perform such simplification: it’s a very easy task, just follow this tutorial.
The next thing you can do is to use the resolution option in Slic3r: it will perform a simplification on the XY plane after the model is sliced. You can set the resolution option to a value like 0.01mm. If you leave resolution to zero, no simplification will happen in Slic3r and the output G-code will reflect the full resolution of the original model.
Finally, you can reduce the number of threads. Due to the way Slic3r is currently written, each threads will reserve its memory space and duplicate most of the geometric data. This means that using 4 threads will reduce the processing time but will use about 4x the memory. You might want to revert to a single thread for very memory-consuming models.
Of course each Slic3r version includes optimizations and things will improve on this side, so expect these hints to be less needed in the future.