Fediverse instances with image hosting are bandwidth limited, but that's just a normal result of image hosting. If you remove image hosting then the bottleneck becomes processing power again.

Sure, but this is largely because currently each client doesn't need to aggregate the whole fediverse. In a decentralised network, you can't split the sum total of processing required to run the fediverse equally amongst peers. Each peer would need to do more or less the same aggregation job, so the total processing required would be exponentially more than with the current setup. You could still argue it's a negligible processing cost per client, but it's certainly way less efficient overall even if we assume perfect i/o etc in the p2p system and even if the client only needs to federate the user selected content

Also just practically deploying a client app that can federate/aggregate constantly in the background (kinda required for full participation) and scale with the growth of fedi without becoming a resource hog I imagine would be pretty tough, like maybe possible yeah but I feel like it makes sense why it isn't like that also