all the knowledge about tor that the NSA has is cointained in those 'slides'
Only the stupid would suggest that, stop being stupid. Tor is well known to not provide defense against GPA / GAA for at least some particular analysis / deployment threats. Some of which are hidden in that doc, including between weasel words, among other threats in other docs by various authors, some of which you've reposted.
if packets arrive at a relay at different times and then, after a delay, all leave the relay at the same time then there's no timing information to exploit.
Yes there is, the packets you sent were observed arriving, or being sent, at different times, in the voids when there was no chaff / fill / other traffic that was not also otherwise discriminated, possibly for even over your whole path through the fully observed network. Harder yes, bulletproof no. If you accepted an unreliable net, where your shit was thrown away due to low traffic conditions, that would be another area of research to do. Depending on use case, unlikely you would accept such unreliability. NNTP flood... you received the msg to you relatively metadata free regarding intended recipient being you due to the flood over wire to disk. But your posts don't receive the same protection as there is no flood there. So you have to use guerrilla or other tactics to post securely, that could pose alternative risks to the poster, including not being able to post [full dataset]. Where are the networks that experiment with fill traffic / chaff? Only thing I've heard is, OMG bandwidth cost, can't do it. Which as said elsewhere is, for many real world applications and users of such a net, bullshit. Dedicate a rate and use that happily.