@amoeba, this is what my anti-PLS friends told me:
Maximum-likelihood PCR outperforms PLS-R (and regular PCR), particularly when measurement errors are heteroskedastic and/or intercorrelated (and when would that ever happen ;) ).
Wentzell, P.D. (2015). The Errors of My Ways: Maximum Likelihood PCA Seventeen Years after Bruce. In B.K. Lavine, S.D. Brown, & K.S. Booksh (Eds.), 40 Years of Chemometrics: From Bruce Kowalski to the Future (ACS Symposium Series, vol. 1199, pp. 31-64). American Chemical Society.