Text is extremely compress-able. Admittedly, this is not common text, and is many "snapshots" of the whole screen, but I believe LZ-based compression would still be extremely effective on this data. The OP above said it can only be one of ten characters, and one of 256 colors, and a 100x40 grid, so it would likely require similar data rates to get this level of quality from a video stream, I would think, since the 10 characters provide more visual information than a simple colored pixel of that size could convey.
The text here may compress well, but only because the original video has been "decompressed" into an incredibly inefficient encoding in the first place. If you want to transmit this efficiently, then downscaling the original video and encoding it efficiently using a standard video codec and rendering it to text client side would be vastly more efficient.
Note you can tunnel arbitrary ports in ssh for data if that's what you're thinking of.