2. Is similar to what I suggest, however, having a rigid cost based on rank denies the system from having cool things like a skill that has a low memory requirement, precisely because of a high skill one, or vice-versa.
It would be boring, and wouldn't take full advantage of the granularity presented by re-scaling the Memory pool.
I disagree with the assertion that having a "fixed cost" for a skill at a specific rank necessitates that you cannot have an interesting variety of options.
You can still have themed progressions, such as a Flare (Basic), Fire Ball (Advanced), Pillar of Flame (Adept), and Lava Fount (Master) skills that each have a variable amount of Memory due to being in four different tiers, but all still serving the general purpose of a "ranged fire skills that ignites the target and potentially the ground beneath them."
The more you invest in either Pyrokinetics OR Memory OR both, the easier it is to include higher Tier Abilities into your line up. You can also choose to have a narrow line up and not expend many points into Memory, but maybe invest heavily into Pyrokinetics, or invest heavily into Memory and only a few points into some other schools to have access to multiple master level abilities from other schools.
The granular aspect would be in the number of Basic/Advanced to Adept to Master skills you decide to slot. Any "odd" number of remaining Memory can be used for a Basic, or most often times, an Advanced ability if you have at least a 4 point investment in the skill's Skill School.
So, you can have those situations in which "man, I really want to use a Master skill, but that will leave me with 5 points left. Lets include one Adept and two more Basic/Advanced skills and we're good."
3. One of the major reasons for shifting towards memory, was to give skills a proper weight when it comes to using them, without stopping you from learning them all.
By eventually making them all cost the same, you eradicate that point in decision making, ending up with the player just picking the strongest ones he has.
Considering that'd be just boring, and is terrible for balance, that's an absolute: No.
If you want to slog through it, you can read more on why I find this idea horribly bad back in the thread already linked.
The issue you point it is a matter of implementation. If you cannot reach realistically "18" in a given Skill School with the absolute min-maxing, then you would never be able to reduce a Master level skill down to the minimum cost of 1.
And, even if this is possible, I honestly don't feel this is imbalanced, as its likely that by investing so extremely in a single skill, you really should be able to use every skill available to that skill tree at once due to all that you are giving up, from better Armor/Health/Magic armor, less options for weapon skills, and a narrower restriction on the other skill schools you can effectively use.
For the builds that don't exclusively pump a single skill school will not reach a point (unless the stats on gear get wildly large later on) in which they can equip Adept skills at "no cost difference" than Basic or Advanced skills in my mock system. You would need at least 8 points in a single Skill School to reduce the cost of the 5 memory Adept skills down to a single memory point, but at this value, it would still leave Master Skills costing 6 memory points, making them hard to squeeze into a build with only 8 Skill Points in the School if you don't also invest some in memory.
This system is also very malleable and can be changed at any time by simply adjusting numerical values in a database, and as such, it would be simpler to balance and change as we and the developers seem fit during EA testing.
My numbers are just a general suggested starting point, but there are several, easy "levers" the developers can fine tune until we hit a sweet spot.
