I tested with C, and got a similar result if I used 'float' while I got a less-rounded result if I used 'double', so I guess perl does it with floats is the reason for the roundoff.

#include <stdio.h>
int main() {
        float d, f;
        d = 0.1;
        for (f = 0.0; f < 7; f += d) {
                printf("f = %f\n", f);
        }
        return 0;
}

$ ./precision-test
...
f = 2.500000
f = 2.600000
f = 2.700000
f = 2.799999
f = 2.899999
...
[download]

(then if I change float to double, it's all 6.900000, etc..)