Solving the ext3 latency problem

Thanks for the explanation.

I guess if i_size could be updated just once, when all the blocks are pushed out, then this would be even less of a problem. But, then again, I have no idea how this actually works inside the code, so this suggestion is probably naive.