From what I know, the information about how to handle no data tiles is needed at layer initialization time, before more tile requests are sent, so that an appropriate handler can be set for every subsequent genuine tile request. If the no data tile were not assessed in advance, more work would need to happen on every genuine tile request, which would obviously impact efficiency. Having said that, I wouldn't expect the initial request to be a lengthy operation, it happens only once per layer initialization, and I would not expect this to surface to the end user in any way including with any perceptible delay in layer initialization - so if you are finding that end users are aware of this specific request in any way (not just from LogCat), then let us know.
Thanks,