Is strong anonymity even theoretically possible on IP based networks?
High latency mixnets (e.g. https://en.m.wikipedia.org/wiki/Anonymous_remailer ) have the drawback that latency means it cannot work with interactive protocols.
Dining cryptographer networks have much lower latency but scale very poorly. (https://en.m.wikipedia.org/wiki/Dining_cryptographers_proble... )
Tor (a low latency mixnet) trades a weaker threat model for low latency and scalability.
So basically you can pick two of scalability, low-latency and resistence to global passive adversaries.
They actually dealt a lot with the traffic analysis problem, and had both a technical and financial model to encourage defense against it. It wasn't perfect, but it would be more resilient to this stuff than Tor IMHO. It just had the disadvantage that latency was atrocious (for obvious reasons), and ultimately it turns out people don't care about anonymity.
For real anonymity you need something that scrambles and delays your traffic to make it harder to track. Something that breaks big transfers up into a bunch of small transfers, sends them via different routes, and generally makes your experience miserably slow.
It would actually be pretty easy to implement for tor (either for the whole network, or individual nodes or routes), but as far as I can see nobody wants to work on it.
https://youtu.be/_2DQ_iYZi5U?t=1580
The tradeoff is you necessarily need to smooth traffic bursts out to meet the fixed rate and that introduces high latency. Unfortunately most user traffic is bursty and not continuous.
Bittorrent has the "break into a bunch of small transfers" part solved. Just need to modify Bittorrent to somehow transfer each piece over a different route.
I2P allows to increase the number of hops to any value.