The problem is due to your setting of the baseline skip, that is too small, so in one case it makes the \lineskip glue to come into action.
Indeed, if you say \vspace*{9pt} before "E. M. Forster" you get almost perfect alignment, as the difference of a bit less than 0.025pt is very tiny; the default value of \lineskip is exactly 1pt.
The descender in "Metamorphosis" is the factor that triggers insertion of \lineskip glue.
Let's do some computations; the depth of the "p" in the title line is 4.87898pt, the height of "Franz Kafka" in the author line is 9.646pt and the sum is 14.52498pt, which is bigger than the current value 14.5 of the baselineskip when the lines are to be set on the page, so the \lineskip glue is inserted.
Indeed, if we try with \fontsize{14}{15} for the author line, the alignment is perfect. However, I'd be more generous with the baseline skips, so to be sure these problems don't appear.

In all cases I've added \par (an empty line is the same) just before \vspace*{8pt}, or there wouldn't be two lines.
Some theory
When TeX breaks paragraphs into lines it stacks the lines one above the other with some glue between them. When a paragraph is finished, another one is typeset and put on the galley in the same way.
The distance from the baseline of a line (where the letters sit) to the following is, normally, regulated by the value of \baselineskip current for the line below (the value can change only when two distinct paragraphs are concerned, because only one value of \baselineskip is used for a single paragraph).
When stacking two lines, TeX computes the depth d of the line above (how much it sticks down the baseline) and the height h of the line below (how much it sticks over the baseline). If b is the value of \baselineskip (for the paragraph the second line belongs to), TeX computes
b - (d + h)
and, if the result is less than the current value of \lineskiplimit (default 0), TeX inserts \lineskip glue. Otherwise the glue inserted is exactly b - (d + h).
Let's assume b = 14.5pt.
In the first case the distance between the baselines will be
d + 1pt + h = 4.87898pt + 1pt + 9.646pt = 15.52498pt
In the second case, the depth of the line above is 0.31448pt and the height of the line below is 9.07199pt, so
b - (d + h) = 14.5pt - 9.38647pt = 5.11353pt
and this will be the glue inserted, making the distance exactly 14.5pt.
To this distance explicit (\vspace) or implicit (\parskip) glues are added. Thus the final distance between the baselines will be 23.52498pt for Kafka and 22.5pt for Forster.
The computations show why using b = 15pt doesn't show the problem.
\vphantom{p}near "Howards End" ? – Lionel MANSUY Feb 05 '13 at 17:32