There is a table in Yoshikowa's paper based upon the representation as graphs with certain markings of vertices. The markings indicate which directions are the A and B smoothings. The table is quite small.

Kamada's book on Surface braids has a nice table of 2-knots that don't have triple points. There are tons of these. So your candidate for small is sort of strange. Braid index might be better.

Quandle cocycles are the only known ways to get lower bounds for triple points, and there are not nearly enough 3-cocycles that have been calculated.

Last spring Dennis Roseman told me a neat way to generate "random" 2-knots. I don't think he has written anything down about it though.

So if you look at movies, you have to be careful about what you define as an event. In the CRS point of view each of Reid. I,II,III, birth, death, saddle, switchback, and psi (pitchfork) moves is an event, but exchanges of distant critical points are also events. Any such event leaves its trace on the chart.

There are other notions of simplicity. For example, there is a notion of thickness. Project into 3-space and subsequently onto the plane (that gives the chart). Take a spear perpendicular to the plane and see how many times it generically passes through the surface. Find the maximum of these, and then minimize over all projections.

Probably the real reason for making a census would be to discover new invariants. So it is a good problem in that sense.