Inconsistent solve behavior

Dear list,

Observe the following behavior.  Why is it that once the denominator
of g becomes not 1, solve seems to hiccup?  Does it do some internal
simplification first?    In particular, is there a flag we should be
setting? (Yes, I realize that to_poly_solve will do these, but it uses
the %union notation, which is superfluous here.)

    (%i2) assume(h>0);
    (%o2)                               [h > 0]
    (%i3) solve(sqrt(x+g)=h,x);
                                           2
    (%o3)                            [x = h  - g]
    (%i4) solve(sqrt(x+1)=h,x);
                                           2
    (%o4)                            [x = h  - 1]
    (%i5) solve(sqrt(x+2.0)=h,x);

    rat: replaced 2.0 by 2/1 = 2.0
                                           2
    (%o5)                            [x = h  - 2]
    (%i6) solve(sqrt(x+1/4)=h,x);
                                   sqrt(4 x + 1)
    (%o6)                         [------------- = h]
                                         2
    (%i7) solve(sqrt(x+2.5)=h,x);

    rat: replaced 2.5 by 5/2 = 2.5
                                   sqrt(2 x + 5)
    (%o7)                         [------------- = h]
                                      sqrt(2)

This was particularly perplexing.  See
http://ask.sagemath.org/question/940/solve-gives-1-2-or-3-answers-depending-if-one
for the original context.


    (%i9) solve(sqrt(4*x+1)/2=h,x);
                                            2
                                         4 h  - 1
    (%o9)                           [x = --------]
                                            4

Thanks for any ideas!  Or I'll file a bug report if this is one - but
the answers aren't *wrong*, so I hesitate to call it a bug per se.