Why roll and not choose? In order to prevent characters from piling up skills way past 3.
That's a very good comment, Omer. Just being Devil's Advocate, though, isn't the threat of being "overskilled" over stated? If the Ref sticks to a character getting about one skill per four game years, that's a hell of a long time for a player to wait.
If a player has Skill-1, let's say we raise him to Skill-2 immediately (similar to the Experience rules), but this raise is not permanent. And, it will take 12 years to get it to Skill-3 (raised to Skill-2 in first term, made permanent in term 2, raised to Skill-3 in third term...averages +1 skill per term).
That's one skill, raised to Skill-3, in 12 game years.
By this time, the character could easily be in his 40's by this time, subject to the critical ageing penalties.
So, although I like your idea, I'm not sure it is practical or necessary from a gaming standpoint.
On Ref Directed Skill Improvement....
I'm all for the Ref making up new methods for characters obtaining skill increases from sleep learning, to university classes, to Psi transfer, to genetic learning by means of a chemically coded shot. But, whatever the method used, I think that the Ref would be prudent to follow the spirit of both the character generation rules and the Experience rules.
More specifically, I think the Ref, regardless of the method, should keep skill improvement to about 1 skill per year (and some variance is welcome).
One method I used to use was copied (not exactly) from MegaTraveller: At the end of each session, I'd have each player put a hash mark next to the skill that was most important for the character for the night's gaming. This hash mark was called an AT or "Action Tally".
I also kept track of each character's birthdate on my Imperial calender. As the ship zoomed off in jump, and through the adventures, I'd keep track of each passing day by checking off days on the calendar.
On a character's birthdate, I'd allow the player to take the skill with the most AT's and make a roll to improve it one level. The roll was a hard one with a base 25% chance of success (thus, the character would improve about 1 skill level every four years, but there is some variance in there for lucky rolls).
Alternatively, if the player wanted a different skill improved, or if he wanted a new skill or to raise an attribute, he could attempt that, but the throw was harder.
Also on the character's birthday, all other ATs are lost, and the process starts again.
In the end, the character gets an improvement about once ever term.
The CT Experience Method....
I like how the CT method works. You get a chance for two immediate skill increases straight out of character generation (if you make the required rolls). You can drop these at any time, then try for something new (IIRC, you've got to wait a year between attempts).
So, you see some fluid skills, going up and down, over the course of a character's career.