Observations on practice on a sample size of 1 with no objectivity, terrible bias, no recorded facts and a complete lack of scientific training ( you have been warned)…
In terms of practice time the answer like everything else is ‘it depends’. Occasional massed practice can be helpful. (reaching)
If you were smart about it you could probably productively improve over time with something like 3 - 15-20 minute sessions per day. With my recent McJob (I recently found a more interesting job) I tended to use commute time. Currently’ish… Review of recently learned material (in this case 2-digit squares - about 70% known as facts) is about 20 minutes. Then practice with 3 digit square in the 100’s which stretches my 2 digit squares for 20 minutes. Then work 2x2, 3x3, or 4x4 multiplication for 20.
If I’m tired/brain dead I may not take on calculation that is hard ‘for me’ and simply work on improving the gaps in my 2x2 squares (currently not terribly difficult at all). …
An extra session on my commute back may help if I’m not tired but may or may not be productive. Number of sleeps often seems to be as important as actual study time in terms of integrating things in a way that becomes a normal part of your thinking patterns. At my age it may never or it may take a solid 10 years (alternately the 10,000 hour rule may be reasonable for expertise - if expertise is integrating a skill to the point that is as natural as language or walking). Gifted folk appear to be able to bypass this volume of effort. In some cases correct training reduces the total effort hours. Incorrect training seems to have the opposite effects.
Some days are sharper than others and I feel like I can stretch a bit further with normal calculation.
I also take days and sometimes a week or more off. I either get stale with practice or side tracked with life. Each time I return there is some back tracking but there is also some new semi-permanent facts that let me move forward a little more quickly and further (not a lot).
There is another aspect to this in that being able to ‘think’/concentrate for longer periods of time is a skill in of itself that appears to take practice. Being a bit obsessive obviously helps but once you reach a point of distraction you are no longer being productive trying to focus narrowly and obsession does not help a bit.
If there were a nice package that monitored learning, calculation speed, strong and weak areas and self-tailored problem solving to the individual I think that learning could be a great deal more efficient. The challenge is that while I could almost certainly get someone else to the point I have reached much faster than I reached it I have no idea how to efficiently move forward other than to use what has worked well so far and read the posts on this forum carefully at the right time, then attempt to extrapolate.
None of the Math-games/trainers that I have found so far do a good job of building a student’s abilities. Most are extremely limited, very few keep track of the student’s abilities, and a lot focus on tricks rather than thinking and integrating facts and methods effectively. At some point I will likely try building an application but in order for it to be really good I think I would/will need a fairly big database of people’s learning statistics in arithmetic calculation that could be leveraged to proactively help individuals improve.
In the interim I keep playing with numbers and trying to get my math skills to a solid undergraduate level while my wife tells me how silly that is.