by making them jump through hoops like sending them directly to willing miners.

Which creates serious collateral damage by making it more profitable to be a large high profile miner.  No one is going to bother tracking down and distributing directly to someone with 1% of the hashpower.   This hurts beyond just the lost income but because mining is always driven towards 0 profit by difficulty adjustment so little cuts can be the difference between breaking even on mining and slowly going bankrupt.

I encourage people to read my reddit comments on the subject: https://old.reddit.com/user/nullc/

The sensible policy is that relay should never be more restrictive than what is reliably getting mined in practice.  Anything more restrictive has the collateral harm of increasing centralization pressure by seriously hurting block propagation performance and by driving transactions to direct miner submission.  Many design decisions in Bitcoin assume that this invariant is largely upheld-- that what nodes relay will match what gets mined.   Even if the transactions are particularly harmful, like slowing nodes down with some resource attack then something must be done (e.g. fixing the resource usage, convincing miners to stop) just continuing to not relay in that case doesn't help and may make matters worse (you want to validate a slow-to-validate transaction in well in advance).

Relay matching mining can, of course, be achieved by miners not allowing these transactions... but they're getting paid millions of dollars to do so, and the protocol assumes that they'll profit maximize more or less up to the limit of consensus rules.  You're not going to convince them to turn away the income and that's a good thing, generally because if they can't/won't resist some angry online mob how could you possibly convince them to resist efforts that were even more persuasive and which targeted transactions you liked?  Filtering out only works to prevent mistakes and casual stupidity, it's a gentleman agreement that kinda worked when Bitcoin was small and more unified but that isn't the world we have today and in most respects we're better off for it.

As d5000 points out, advocacy on this point is misdirected-- this is a less harmful means while much more harmful means are (1) currently in use and no less attractive, (2) impossible to block.  So the extent that some of these users may intentionally be attacking rather than just idiotic, making sure that the least damaging avenue is available helps make their intentions more clear.

Moreover, the argument that the filtering isn't censorship or at least a means that would work equally well for censorship only works because the filtering doesn't work.  Defending something that is only not bad because it doesn't work seems like a total folly to me.  And ultimately it gives the activity free press, which for some may even  be the primary reason they were transacting in a wasteful way to begin with.  Over and over again these activities have gone away when people ignore them.

The spam sucks, but Bitcoin is already designed to handle it, it's one of the reasons that there is and must be some block capacity limit. That's (part of) what it's there for, and it support's bitcoin's core value of minimizing subjective human influence on the ways that third parties can interact.  It's always the case that you could make something better by injecting a good human judgement call,  but in doing that you take the risk of a bad human judgement call.  Better to minimize human judgement, and set crisp content neutral boundaries at the points where some decisions must be made.