With their data replicated over so many servers, I don't think Google place the same emphasis on data integrity as us mere mortals. Their server environment is unique and it is difficult to draw conclusions from their decision.
I would still be wary of ext4 for the moment for a production server, its reputation and novelty remain stumbling blocks to adoption. Give it another year or two to mature, and we'll see if stories regarding its tendency to lose data is undeserved. If ext3 works, then why jump too fast?
That proves encyclo's point: they are far more interested in performance than data integrity at the file system level because they get data integrity at the GFS level.
I intend to stick to ext3 for a while, and then may not move to ext4 - someone is working on ZFS on Linux (via FUSE), then thre is BTRFS. IN a few years we might be all using flash drives and there are specialist file systems to get the best performance out of those.