Here's code for the "cap on separate para" version:
Code:
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<title></title>
<style>
p {margin:0; padding:0;}
.para-dropcap {
text-align: center;
font-size: 4.6em;
font-weight: bold;
line-height: 1;
}
.para-chapterFirst {
text-indent: 54.5%;
text-align: justify;
margin-top: -1.75em;
margin-bottom: 0.1em;
}
</style>
</head>
<body>
<p class="para-dropcap">E</p>
<p class="para-chapterFirst">nter all ye who would ... yadda yadda
yadda yadda yadda yadda yadda yadda yadda yadda yadda yadda yadda
yadda yadda yadda yadda yadda</p>
<p class="para-dropcap">W</p>
<p class="para-chapterFirst">est of ... yadda yadda
yadda yadda yadda yadda yadda yadda yadda yadda yadda yadda yadda
yadda yadda yadda yadda yadda</p>
<p class="para-dropcap">T</p>
<p class="para-chapterFirst">esting the centering ... yadda yadda
yadda yadda yadda yadda yadda yadda yadda yadda yadda yadda yadda
yadda yadda yadda yadda yadda</p>
<p class="para-dropcap">I</p>
<p class="para-chapterFirst">n the still ... yadda yadda
yadda yadda yadda yadda yadda yadda yadda yadda yadda yadda yadda
yadda yadda yadda yadda yadda</p>
</body>
</html>
You can see that the "E" looks good, the "W" is OK, the "T" is bad, and the "I" is terrible.
Since this causes text-to-speech to break while not completely fixing the formatting, I picked the way that kept TTS functional.