I don't think there is any reason to make block *expensive*, per se, but if the threshold is too low then blocks will be generated in rapid succession, with negative effects on synchronization between peers. To perform useful work the client must know about the most recent block in the chain, since a reference to the previous block is among the criteria for accepting a new one. Currently one block is generated every 10 minutes (ideally). Increase that by much more than a factor of two or three and many of the clients may end up wasting most or all of their efforts on outdated block chains.
Worse, those who generate the most blocks within a single private network, with their peer connections most likely routed through a single server, will tend to get updated most quickly when new blocks are found. This lets them start work on the new chain sooner, and thus use their CPU time more effectively. Speeding up block generation would tend to give them even more of an edge in that regard.