Oh, correlation is not causation?
The crap in that article you referred to is based on models. Everything is hypothetical.
If the state of modeling is sufficient to do the atmosphere over millenia, why can't you tell me what Microsoft shares will trade for tomorrow?
According to your article, the total worst case effect on CO2 solubility for the widest salinity range would be 6.5% of the observed (correlated) CO2 concentration. In fact on average it would be considerably less.
Yes, the article is based on models -- models in which the fundamental parameters are constrained by observational data, and the results must be compared to observational data. One of the issues I'll touch on for point #5 is that the best models don't yet account for the full range of CO2 variation in glacial-interglacial cycles; so far, the best that's been done is about 2/3 of the total. That tells the researchers they don't get it yet -- but they clearly do get that the effect of warming or cooling the ocean is not the major factor.
So, to answer your question in a straightforward way, these are not predictive models.
I pointed out the salinity aspect because it's a basic part of the system, yet one not even mentioned by the author of the piece you linked to. The authors of the paper I linked to are addressing everything, EVEN factors that are only responsible for less than 10% of the total, because that's the only way to do it correctly.