Pyrex: thanks for the responses.

And yes, you were interpreting my words correctly.
Essentially, I think we may have to resolve it to playtesting and/or agree to disagree, because I believe that giving a caster the ability to cast more high level spells at the cost of casting a total number of spells overall is ok. It seems perfectly fair to me to give up casting 3 or 4
burning hands in favor of one
fireball. Sorcerers - to continue to use this one class as an example - are all about burst damage (not area-of-effect necessarily, but high damage output per round), and giving them more
fireballs to throw (or not) gives them more options. Other spontaneous casters generally follow suit. And giving a wizard the ability to cast anything he knows? That's darn near priceless.
Also, I do have to take issue with the 17th level wizard with 19 Int. It's just not a fair counter-example for the problem; a 17th level wizard, even a pretty poor one, will most likely have over 24 Int - which would be a significant gain in mana points. The way I calculated examples for this system myself were: level 1 with 16 in stat, level 6 with 18 in stat, level 10 with 22 in stat, level 17 (or 18) with 24 in stat, level 20 with 26 in stat. These numbers are really not unreasonable, and they show a better picture. (In fact, min/maxers will have closer to 30+ by level 20.) And I will certainly admit that this system does give a greater emphasis on +stat items to get those extra spells - it's just part of how it works.
And, most importantly, it's exceedingly difficult to give the sorcerer any more spell points without making the system unbalanced. If a sorcerer is allowed to cast (for example) 10 level 9 spells at level 17, who cares how many level 1, 2, 3, 4, 5, 6, or 7 spells he can cast? He's just
gated in 10 angels or rained down
meteor swarms for a full minute. He
wins.
It's a hard line to draw, I admit - but I still think it's balanced.
All that said:
What
really needs to be developed - and I just can't think of a good way to do this without getting more complex than it needs to be - is to come up with a system that allows low-level spells to cost
less when higher level spells are available. Something along the lines of: when you are able to cast 5th level spells, 1st level spell costs drop by 1/2. This would allow folks to continue to spam the "tiny" spells while not giving them access to too many "big" spells. If someone could come up with a smooth system for doing this, it would help solve your issue while maintaining mine. How's this for a starting point, using my system's rules above:
Each time a caster has access to a higher spell level, all spells 3 levels lower or less cost 1 mana point fewer to cast, to a minimum of 1 point. For example, once a wizard is able to cast 4th level spells, all 1st level spells cost 1 point. Zero-level spells do not change because they already cost just 1 point. When the same wizard is able to cast 8th level spells, 5th level spells will cost 5 points (1 fewer), 4th level spells will cost 3 points (2 fewer), and zero through 3rd level spells will cost 1 point (3rd level spells are 3 fewer, and the rest were all just 1 point by then). These point reductions do not affect the base cost of adding metamagic to spells, but they would apply to the final cost of a spell affected by metamagic. For example, an 18th level sorcerer could cast an empowered scorching ray for 1 mana point (2 points + 2 normally, then adjusted downward as all other 4th level spells).
By this rule, when a wizard is level 17, 4th level spells are just 2 points - and only 1 point for the sorcerer. This may be TOO generous (I've done scant few calculations), but it's starting out in the right direction, I think... (It's also starting to get pretty confusing, especially with metamagic...)
(However, I still like the idea of "small pool + rapid regen" idea... perhaps another thread is in order...

)
HeavenShallBurn: One pitfall that needs to be avoided when making mana point systems is to remember that zero-level spells cannot cost nothing; if this is the case, then a
cure minor-casting cleric will simply spend 10 minutes to heal a party completely after each battle, and no one would ever buy a torch again if they had a wizard (or cleric) in their party with a
light spell.