That website is mistaken, or rather it's giving a simplified explanation that will mislead you in many cases.
As ssav pointed out, there are tons of counterexamples to that rule when it's stated as a universal principle. Some of these counterexamples follow patterns of their own which you may be able to learn; others are unpredictable violations that pretty much just have to be memorized one by one.
I found the following document that gives a more detailed and technical overview of stress in words ending in -al: Is the Adjectival Suffix -al a Strong Suffix?, by Quentin Dabouis.
Usually stressed on the penult
Adjectives ending in -C.Cal, where C.C is a consonant cluster that can't come at the start of a word
The most important class of counterexamples: if an adjective ends in -al preceded by a consonant cluster that can't come at the start of a word (including a pair of identical consonant letters), it is likely to be stressed on the penult.
This is the case for all of ssav's examples (rebuttal, universal, accidental, infernal, horizontal, fraternal). The only exception I know of is sagittal, which is often pronounced with antepenult stress, although it can also be pronounced with penult stress.
Adjectives ending in -al derived from nouns with antepenult stress
Adjectives with more than three syllables may fall into another class of counterexamples. According to John Wells, if you take a noun with antepenultimate stress and derive an adjective ending in -al, the stress on the adjective is likely to be on the penult due to the Alternating Stress Rule. An example of this: the noun hómicide corresponds to the adjective homicídal. I think this explains why from ádjective we get adjectíval and from súbstantive we get substantíval.
Usually stressed on the antepenult
Antepenult stress is only the rule for some cases.
Adjectives ending in -ical
For example, it is generally true that adjectives receive antepenult stress if they end in '-ical.' You'll notice this explains the two words listed on that website, nautical and critical. (The only possible exception I can think of is cervical, which actually can be stressed on either the antepenult or the penult; I think antepenult stress is more common nowadays. The reason it can be pronounced differently is probably because, etymologically, it's from the stem of "cervix" + "-al" rather than having the "-ic" suffix like other words ending in "-ical.")
Trisyllabic adjectives ending in -VCal
It's also generally the case that trisyllabic adjectives that end in VCal (a single vowel letter, followed by a single consonant letter that isn't x, followed by the suffix -al) are stressed on the antepenult (the first syllable). For example, radical, doctrinal, intestinal, festival, personal are all stressed on the antepenult.
However, even among this class of adjectives, there are some that may optionally receive penult stress for etymological reasons relating to vowel length in Latin. (John Wells discusses this in more detail in the linked blog post.)
In addition to the aforementioned cervical, all of the following trisyllabic adjectives have two pronunciations, one with stress on the antepenult and one with stress on the penult: vaginal, coronal, palatal, urinal, archival, gingival. In some cases, one pronunciation is more common than another or associated with a specific region or profession.
Also note this rule does not apply to trisyllabic nouns ending in -VCal, of which there are several with penultimate stress such as revival, survival, arrival.
Quentin Dabouis, « Is the Adjectival Suffix -al a Strong Suffix? », Anglophonia [En ligne], 21 | 2016, mis en ligne le 01 juillet 2016, consulté le 19 août 2017. URL : http://anglophonia.revues.org/754 ; DOI : 10.4000/anglophonia.754
John Wells, "skeletal", John Wells's phonetic blog