I agree that the common explanation of the longitude of periapsis, as the question gave it, appears at first sight to make no sense -- because it constructs a sum of two angles in two different planes, and that seems at first sight to lack geometrical meaning.
In fact the fault, if any, is in the explanation rather than in the geometry. It is not so difficult to show a geometrically sensible construction for the longitude of periapsis, without adding any angles in different planes. But before showing this, it may be worth answering the question 'Why do it at all, what's the point?'.
At this point the present answer links up with the answer given by James K. One of the uses of such a construction came to attention in the early days of astrodynamics. When programmed orbit calculations are to be done for artificial space objects, the orbits to be calculated can in principle have any values whatever of the inclination and eccentricity, in principle down to zero. When either or both of these orbital quantities approaches zero, then the position of the node and/or the periapsis, or else computations involving them, can become in practice numerically ill-defined, or anyhow show some undesired numerical behavior. As a result of the difficulty, searches were made a long time ago now, for alternative systems of orbital elements that do not generate undesired numerical behavior at zero or near-zero values of inclination or eccentricity.
The resulting formulae for orbital equations, usable for low-inclination, low-eccentricity orbits, typically made use of devices that included the equivalent of replacing the anomaly of an orbiting object by its longitude from a defined initial point, and replacing the argument of periapsis by a corresponding longitude from the initial point. The development of orbital equations in terms of such quantities can usefully be generated, or at least illustrated, by the following construction that also provides an in-plane account of the longitude of periapsis, the subject of the question.
Sources for the geometry and formulae connected with this kind of development include R M L Baker (1967) 'Astrodynamics', especially Appendix C, and applications are shown for example in A E Roy, 'Orbital Motion' (especially 4th ed., 2005) (or google for other sources of the same title), especially chapter 8, Special Perturbations, section 8.5, for perturbational equations and their derivation.
Baker's 'Astrodynamics' illustrates the definition for these purposes of an alternative ('F G W') coordinate system. See the figure below.
One starts with ordinary rectangular coordinate axes defined by points X, Y, Z on a unit sphere with center O, where the xy plane is for example the ecliptic plane, and OX is in a chosen initial direction, e.g. towards the equinox of date. Define an orbital plane through F N G, intersecting the xy-plane at node N with inclination angle $i$ . Angle XON is then the longitude $\Omega$ of node N. Define axis OW perpendicular to orbital plane F N G. Let angle FON = XON, and let angle FOG = 90°. Then make F the initial point of coordinate system FGW. In this coordinate system, the longitude of the orbital periapsis starting from the initial point is FOP, which is the sum, as usual, of $\Omega$ the longitude of the node FON and $\omega$ the argument NOP of periapsis, but now they are measured in the same plane -- which was the thing to be illustrated.
This FGW system can be conveniently oriented by the coordinates of point W, the pole of the orbital plane,
which has coordinates (in terms of the x,y,z axes and the quantities $i$ and $\Omega$) of
$ W_x = + \sin (\Omega) . \sin (i) $ ,
$ W_y = - \cos (\Omega) . \sin (i) $ ,
$ W_z = \cos (i) $ .
Then initial point F can be shown (with no difficulty but at a little length) to have coordinates relative to initial axes x, y, z, of
$ F_x = 1 - W_x^2 / (1 + W_z) $ ,
$ F_y = - W_x W_y / (1 + W_z) $ ,
$ F_z = - W_x $ .
Quantities Wx, Wy, Wz are sometimes identified as $h_x/h_m, h_y/h_m, h_z/h_m,$ and can be called direction cosines made by the orbital angular momentum vector (in direction OW) with axes x, y, z.
You might also find some of the formulae useful in Preliminary orbit-determination method having no co-planar singularity by R M L Baker and N H Jacoby, Celestial Mechanics 15 (1977) 137-160.
