If mining is centralized then the exit points are probably easy to find.
Can you please explain this in more detail? Neither are there exit nodes, nor are transactions sent directly to miners. Transactions are broadcast through the whole network. How do known miners make it easier to perform traffic analysis in a cover traffic scenario?
I assume you mean relay nodes sending out dummy packets at random intervals, so latency doesn't increase for legitimate traffic (as long as relay nodes can handle the additional bandwidth).
I have been thinking about establishing fixed capacity channels between sets of nodes instead. The negotiated capacity is filled completely at all times, either with padding or real data. Because channels are encrypted, an attacker cannot differentiate between them.
Seems to me the adversary could ignore all packets coming out of the exit nodes that didn't correlate with a low-latency to the targeted entry node.
Again: There are no exit nodes with I2P. This is also not possible with fixed bandwidth cover traffic channels.
Sybil attacks are very hard to defend against if your attacker can replace your internet connection with connection to NSA LAN instead. Despite all, in 2012 the NSA was still obstructed in some degree by even Tor use. The least we can do is make it a bit harder for them.