Did you mean to file this as a bug report for the docs?
Seems like this is a problem with the extension rather then the docs?

My bad - I did not mean for this to be a bug against the documentation. Thanks for redirecting it.

It was brought to my attention that I did not correctly describe the issue. Here is a more accurate summary of what the problem is: 

The code in php (libgd) uses libvpx and writes the riff manually actually. The code generates the correct even size, but neglects the padding. It's possible older versions of libwebp would decode this, but libwebp 0.4.0 does not.

Here is a proposed fix from one of the WebP developers with the disclaimer that it is untested and may have incorrect local style.

 diff --git a/ext/gd/libgd/webpimg.c b/ext/gd/libgd/webpimg.c
index 01bef93..ca4e9bc 100644
--- a/ext/gd/libgd/webpimg.c
+++ b/ext/gd/libgd/webpimg.c
@@ -778,6 +778,18 @@ WebPResult WebPEncode(const uint8* Y,
 										(chunk_size >> 16) & 255,
 										(chunk_size >> 24) & 255 };
 	  memcpy(*p_out, kRiffHeader, kRiffHeaderSize);
+	  if (img_size_bytes & 1) {  /* write a padding byte */
+		const int new_size = *p_out_size_bytes + 1;
+		unsigned char* p = (unsigned char*)realloc(*p_out, new_size);
+		if (p == NULL) {
+		  free(*p_out);
+		  *p_out = NULL;
+		  *p_out_size_bytes = 0;
+		  return webp_failure;
+		}
+		p[new_size - 1] = 0;
+		*p_out_size_bytes = new_size;
+	  }
 
 	  if (psnr) {
 		*psnr = WebPGetPSNR(Y, U, V, *p_out, *p_out_size_bytes);

Indeed, I can confirm this issue. Consider the following simple
test script:

<?php
$im = imagecreatetruecolor(75, 75);
$red = imagecolorallocate($im, 255, 0, 0);
imagefilledrectangle($im, 0, 0, 74, 74, $red);
imagewebp($im, __DIR__ . '/bug66590.webp');
?>

This fails to create a valid Webp image file. display (ImageMagick
6.9.1-2), for instance, reports: "insufficient image data in file
[...]". The length in the RIFF chunk is given as 92 bytes, so the
file size should be 100 bytes (8 bytes RIFF header size), but it's
only 99 bytes.

The supplied patch obviously fixes this issue. Thanks!

Reported upstream against libgd:
<https://github.com/libgd/libgd/issues/176>.