The connection between the escape fraction of ionizing photons ($f_{\rm esc}$) and star-formation rate surface density ($\Sigma_{\rm SFR}$) is a key input for reionization models, but remains untested at high redshift. We analyse 35 z~3 galaxies from the Keck Lyman Continuum Survey (KLCS) covered by deep, rest far-UV spectra of the Lyman continuum (LyC) and high-resolution HST V$_{606}$ imaging, enabling estimates of both $f_{\rm esc}$ and rest-UV sizes. Using S\'ersic profile fits to HST images and spectral-energy distribution fits to multi-band photometry, we measure effective sizes and star-formation rates for the galaxies in our sample, and separate the sample into two bins of $\Sigma_{\rm SFR}$. Based on composite spectra, we estimate <$f_{\rm esc}$> for both $\Sigma_{\rm SFR}$ subsamples, finding no significant difference in <$f_{\rm esc}$> between the two. To test the representativeness of the KLCS HST sample and the robustness of this result, we attempt to recover the well-established correlation between $f_{\rm esc}$ and Ly$\alpha$ equivalent width. This correlation is not significant within the KLCS HST sample, indicating that the sample is not sufficient for correlating $f_{\rm esc}$ and galaxy properties such as $\Sigma_{\rm SFR}$. We perform stacking simulations using the KLCS parent sample to determine the optimal sample size for robust probes of the $f_{\rm esc}$-$\Sigma_{\rm SFR}$ connection to inform future HST LyC observing programs. For a program with a selection independent of ionizing properties, >= 90 objects are required; for one preferentially observing strongly-leaking LyC sources, >= 58 objects are required. More generally, measuring the connection between $f_{\rm esc}$ and $\Sigma_{\rm SFR}$ requires a larger, representative sample spanning a wide dynamic range in galaxies properties such as $\Sigma_{\rm SFR}$.
Comment: 13 pages, 5 figures, accepted for publication in MNRAS