The **Hairiness Conjecture**, formulated by Milnor and proved by Lyubich ("Feigenbaum-Coullet-Tresser universality and Milnor’s Hairiness Conjecture", Annals of Mathematics, 1999) states that, near any real Feigenbaum parameter, the rescalings of the Mandelbrot set converge to the whole complex plane in the Hausdorff metric.

This means that, if you fix any resolution that you want, then at a sufficiently deep zoom near such a point the distance-estimate method will show you only black pixels.

**EDIT.** Let me elaborate also on Alex's answer (since this is going to be too long for a comment), and the difference between the Misiurewicz and Feigenbaum parameters.

As Alex points out, suppose that you fix your resolution, say 5000x5000 pixels. For simplicity, let us assume that our picture is always square, so that the resolution is determined by a single natural number, N, e.g. N=5000.

Then Alex's point is that you can find a Misiurewicz parameters $c_0$ such that, for sufficiently small magnification, the picture drawn around $c_0$ at resolution N by N contains only black pixels. Or, more mathematically: For all sufficiently small $\varepsilon$, if $c$ is such that $|c-c_0|<\varepsilon$, then
$$\operatorname{dist}(c,\mathcal{M}) \leq \varepsilon/N \tag{*}$$.

If you know a little bit about the Mandelbrot set, this is quite obvious: take one of the "star" Misiurewicz parameters, which are at the end of a bulb of period $q$ bifurcating from the main cardioid, and which have $q$ "arms" attached to them. These arms are essentially equally spaced out (this follows e.g. from the picture in the dynamical plane, which is asymptotically self-similar, and which is repeated in the parameter plane by a theorem of Tan Lei). If the number $q$ is much bigger than the number $N$, then the claim follows.

On the other hand, if you **fix a Misiurewicz parameter $c_0$ first**, and then choose $N$ sufficiently large, you will always see something of the complement of the Mandelbrot set, no matter how closely you zoom in. (This is essentially for the same reason.)

The hairiness theorem implies that, for any real Feigenbaum parameter, this is not the case: No matter which resolution $N$ you choose, (*) will hold when $\varepsilon$ is sufficiently small.

For an even easier (though less direct) argument - again using a nontrivial theorem - recall that Shishikura proved that **the boundary of the Mandelbrot set has Hausdorff dimension two**. In particular, its box-counting dimension is two, which immediately implies the desired property ...

Moreover, the set of parameters at which the dimension is locally two is topologically generic in the boundary Mandelbrot set, so zooms around a generic boundary parameter will result in the desired picture.