In addition to Redeemer's post, the reason why I said that you should use small(er) blocks is simple:
Take a small block (Add -> Add Cube -> Small). The shaded block will now consume a little amount of video memory (for the shadow map). When you now scale the block, f.i. about a factor of 100, and keep the default lightmap scale (Map Compiler settings), the block will take a hundredfold of video memory.