I should clarify more on that point. The compression patent says that up to 255 x 255 (~65k) image blocks may be searched for each block to be coded. Considering that an HD frame can have over 130k 4x4 blocks, that is a lot of time spent on block matching. However, in practice the searches are faster if a suitable block can be found early, and if a large maximum block size is used. But it still takes a long time, especially if a frame has a lot of detail, because this will cause large blocks to be split into four smaller blocks. The patent says that it is easy to arrange compressing different sequences of contiguous p-frames on their own threads, so a multicore PC would be useful.
So it is not something you would want to run on a normal desktop. I can see why TMM made the server farm. You really need many computers in order to make it practical. I also do not see how the compressor can ever work in realtime, even though the patent claims it.