New blocks will be mined at the edge of the accepted time, this will lead to many orphan blocks and network forks (probably short). Maybe more problems will arise, comparison with Bitcoin isn't applicable because in Bitcoin there is no reason for miners to set time to higher values and they just set it to current time and whole network accepts their blocks.
Blocks at the edge of accepted time will not be used by miners as base. Miners will be interested to mine starting from fastest blocks to be able to put their blocks in longer chain. Mining on already short chain pointing to the future is not practical. The same is with bitcoin now. The only real winning strategy is to use blocks who pointing to a moment in time which is closest to the genesis block among chains of the same length.
But the idea of a map enlargement looks also very interesting. This is very easy to implement right now and no need in such a deep analysis and looking for a pitfall. The map is not included into block and will not enlarge it. The size of a map will not affect emulation speed because collision detector needs to check only few closest blocks on every frame. And this solution will automatically kick off any map preselection cheaters. This is a very smooth way to go. Really. Lets remove the time threshold and let the map size be a subject for retarget.
I would be keen to see an experiment with this.