Yes this is exactly true, proteins are ideal for deep molecular dating. DNA can suffer from compositional bias. The thing you need is a reliable tree, with strong bootstrap support. You will then calibrate it with your externally established dates.
Programs such as Beast allow you to do the two simultaneously. Generally, whatever approach you take you will be required or else requested to use Beast. It is however impressive.
What are you dating? Ten million years is not 'deep', however 300 million is very deep indeed for animals (Metazoa), especially so for vertebrates.
To answer your question in general terms molecular dating tends to estimate earlier divergence against the fossil record, or against the perceived view of divergence. Molecular phylogenetics argue this is because the fossil record is not sufficiently well sampled, ie the earliest fossil is missing. Palaeontology historically had a different viewpoint.
The plethora of molecular dating methods has continued, an impressive Beast based approach was recently published and I'm not clear how these changes/ improve the dating estimate. My personal view is that generally each new technique improves the accuracy of molecular dating.
Answering the second part of your question, reconstructing the ancestral sequence at a given node is formal analysis and can be performed using ML, Bayes or parsimoney. Thus it attempts to estimate the ancestral sequence and you can then compare the mutations against extant taxa. It is perhaps better to say it establishes a robust hypothesis regarding ancestral mutations.