Generally, it's a bad idea to have data of different levels of granularity held in the same table. Additionally (like other statistical software) JMP requires data in a column to be of the same data type, modeling type and format.
But, so long as you don't mind a proportion rather than a percentage, if you want to do this you could use JSL:
NamesDefaultToHere(1);
// (1) Make some data
nr = 10; // Number of rows
nc = 5; // Number of columns
dt = NewTable("Data");
for (c = 1, c<=nc, c++, dt << NewColumn("Col "||Char(c), Formula(RandomINteger(-10, 10))));
dt << addRows(nr);
dt << runFormulas;
for (c = 1, c<=nc, c++, Column(dt, c) << deleteFormula);
// (2) Insert a new first row
dt << addRows(1, atStart);
// (3) Populate values of first row
for (c=1, c<=NCols(dt), c++,
vals = Column(dt, c) << getValues;
vals = vals[2::NRows(dt)];
Column(dt, c)[1] = NRows(Loc(vals < 0))/(NRows(dt)-1);
);