1. It is worth noting that the speed of distribution of blocks in general on the network does not play a special role for mining. The main thing is what is the speed of distribution of blocks within the subnet that covers miners. If the speed inside the subnet was at the level of 1 second, we would have ~1.6% of orphan blocks. But there are much fewer orphan blocks at the moment. Somewhere on the forum, a figure of 0.0024% was given (I can't vouch for the accuracy). This means that the speed is now much less than 1 sec.
Correct. Keep in mind that doing so would make the current mining scene more centralized, whereby miners that are not interconnected with the majority of the miners are more likely to experience more orphans, due to the lack of a direct connection.
2. When using a technology such as spv mining, the block size ceases to affect the number of orphan blocks. Miners start mining the next block almost immediately after the block appears. However, this increases the number of empty blocks. But I think this is quite a reasonable compromise: reducing the number of orphan blocks at the expense of some increase in empty blocks. Moreover, an empty block, which sometimes appears a few seconds after a non-empty block, does not cause any inconvenience to users.
From all this, we can conclude that, for example, 10-20 MB. blocks do not pose any threat to the security of the network in terms of orphan blocks.
I would prefer miners to validate the blocks first. SPV mining is not a behavior that I think we should encourage, we've had a mishap with SPV mining in the past so probably not such a good idea given that current validation speeds are sufficiently fast.
I don't think those sizes are particularly unrealistic to achieve in the future. If you don't need the majority of the network to be able to run their own nodes and that mining to be centralized with a few entities, then it should probably be fine.