But there are plenty of service jobs that haven't seen big productivity gains. Masseuses, for example.
I'd argue this is still a version of the Baumol effect. For example, some people look back at 1950s nightclubs and wonder why they can't have that now.
It's because, as you say, the productivity of automatic music increased but live orchestras didn't.
So if you want "music" you can get it cheap, if you want "live orchestra", you can't.
People were able to afford something earlier which in these days they no longer can.
https://www.aaup.org/sites/default/files/AAUP-2021-SurveyTab...
Baumol's cost disease theory makes sense in general. But empirically, for a specific service, there's always a lot of other stuff going on, are they providing a superior good with rising demand as society gets richer, how easy is it to substitute lower quality adjuncts, etc.
In 1900 a performance could entertain at most a few thousand people. Now, a single performance can be heard by tens of millions. A performance in the Concertgebouw Amsterdam can be heard in real time* in New Zealand, on the other side of the world, with better fidelity than all but a few of those present in the theatre experience.
Baumol's choice of example of zero-productivity-gain work was egregiously wrong in this case. Likely the same is true for his other examples.
* Or later, at the listener's convenience.
So if the orchestra's economic production is still only being realized when they perform live music, the baumol effect will absolutely still apply.
Now, yes: internet streaming killed recording as a viable profit center in the mid 2000s. When Baumol was writing, musical acts did live shows more or less at cost, or even at a loss (covered by their record companies), to promote their recordings, which were where the money was, then.
The declining popularity of classical music is beside the point. Baumol intended classical music to stand in for performing arts in general.
The size of the potential (and in many cases actual) audience and revenue has increased for performances of all kinds: popular music, stand-up comedy, baseball games, motor car races, spelling competitions, etc. At the same time, costs to an individual consumer have declined dramatically. Those things are the essence of productivity gain.
Economics feeds into this. I happen to like (among other things) "big band" music, which flourished in the 30s through the end of WWII. The reason was that wages were depressed (it was The Depression, after all) so it was cheap to hire a big band (and it was a way to spread the money around and keep your friends employed). The war ended, the depression ended, and jobs were plentiful (plus some new technology wsa developed) and that was the end of the Big Band Sound.
Symphonies evolved in a time of patronage and low wages. They are absurd today -- real money losers. I'm glad people still try, but I feel like my experience must be subsidized.
I am compulsively unable to attend a concert without estimating the amount the band is paid and dividing it by the players. The economics rarely make direct sense.
The second is that college professors have been phased out of teaching for a long time, increasingly replaced by adjuncts who earn starvation wages.
Healthcare, I'm not sure about. At my most recent clinic visits, a doctor (i.e., MD) was nowhere in sight. A technician examined me, a physician's assistant interpreted the exam results, and ordered a visit to a MRI machine operated by technicians. (Thankfully, it turns out I'm OK). My healthcare costs twice as much as in most civilized countries for no good reason despite roughly equal labor inputs.
It would be useful to remove education and healthcare from the graph and see if the theory still makes sense.