BACKGROUND
Suppose we have a journey between $A$ and $B$ where different sections of roads have different speed limits and our average speed is $50$ mph.
If when I repeat the journey I travel $10\%$ faster at each point, then clearly my average speed will be $10\%$ faster and I will arrive in $1/1.1$ of the original journey time.
On the other hand, if I travel $5$ mph faster at each point, then the average speed may be $10\%$ faster, or it may be different. (For example, if there is a incredibly slow section of road with a limit of $1$ mph, then it takes ages to go along it at $1.1$ mph, but hardly any time at all when I go at $1+5$ mph.)

Question
When the average speed is $50$ mph, does the option of going $5$ mph faster always beat (or equal) the going $10\%$ faster option? If so, is there a simple proof? (This is just a question I was pondering while in a car so I don't know the answer or even whether this is a trivial or hard question...)
What I've tried
I've written a simple Python program to simulate this and experiments indicate that it is always better to go $5$ mph faster (unless all speeds are identical in which case it makes no difference.)
To try and prove it I imagined that there were $50$ sections of road, each $1$ mile long with speed limit $v_i$. As the average speed is $50$ mph, I know that the total time is $1$ hour and so:
$$ \sum_{i=1}^{50} { 1 \over{v_i} } = 1 $$
I wanted to show:
$$ \sum_{i=1}^{50} { 1 \over{v_i + 5} } <= {10 \over 11} $$
On wikipedia I found that the harmonic mean is less than the arithmetic mean which felt useful, but my attempts to use this got nowhere.
Can anyone find a counter example or perhaps point me in the right direction?
When distance is fixed and time is variable, it is often more natural to work with the reciprocal of speed, which I'll call "slowness" and denote $w=1/v$. I'll state a few easily verified properties of slowness:
We also have $\int_0^X\mathrm d\mu=1$, which will be needed later.
The problem is to compare two possible journeys: (I) where you increase your speed to $v'=kv$, and (II) where you increase your speed to $v''=v+\delta$, with the condition that $k\bar v=\bar v+\delta$.
In (I), your average slowness becomes $\bar w'=\bar w/k$.
In (II), your slowness becomes $$w'' = g(w) = \frac1{1/w+\delta},$$ which you can verify is a concave function of $w$. Jensen's inequality therefore implies that $$g\left(\int_0^X w\,\mathrm d\mu\right)\ge\int_0^X g(w)\,\mathrm d\mu.$$ The left-hand side is $g(\bar w)=\bar w'$, the average slowness of (I). The right-hand side is $\int_0^X w''\,\mathrm d\mu=\bar w''$, the average slowness of (II).
Therefore, (I) is at least as slow as (II), with equality holding only if $w$ is constant almost everywhere.