Unfortunately I don't think I should implement this because it would involve a lot of special handling that would only benefit some pretty unusual use cases. The panel_axes infrastructure is built around the assumption that panels represent some reduced dimensionality of the main subplot data, so it usually doesn't make sense to have a map projection panel.

I think what would make more sense in your case might be to make the Hovmoller plot the "panel" and make the projection the main subplot. You can make panels as large as you want -- nothing stopping you from having a narrow main subplot and a much larger panel. Here's an example (development version is needed to pass format keyword args to panel()):

import proplot as pplt
slon=0
elon=360
proj = pplt.Proj('cyl', lon_0=180)
fig = pplt.figure()
ax = fig.subplot(
    proj=proj, land=True, abc='A.', title='Title', titleloc='l',
    latlim=(25, 85), lonlim=(slon, elon), labels=True, lonlabels='t'
)
px = ax.panel(
    'bottom', width=2, xformatter='deglon', share=False,
    xlim=(slon, elon), xlocator=60, number=2
)
N = M = 100
x = np.linspace(0, 360, N)
y = np.linspace(0, 25, M)
z = np.sin(x[:, None] / N + 20 * y[None, :] / N)
px.contourf(x, y, z, cmap='viridis')

Also I realize one limitation of this is that you can't put colorbars along (e.g.) the left or right side of the Hovmoller plot, since it's a "panel".... but I think the generalization proposed in #329 will add support for this.